Post-processing for Thorax Imaging based on Electrical Impedance Tomography using Deep Learning

Electrical Impedance Tomography (EIT) for lung monitoring is a non-ionizing medical imaging technique that is on the rise in biomedical engineering research in the recent decades. It reflects the inner conductivity distribution of biological tissue based on boundary voltages that appear on the electrodes, after injecting an alternating current of a certain amplitude into the human thorax. EIT, however, has a low spatial resolution and is susceptible to errors caused by electrode movement and poor contact quality, which can lead to the generation of blurry images. Post-processing imaging techniques can, in fact, be a beneficial tool for fine-tuning this issue. The focus of this work is to implement a U-NET multi-class semantic segmentation deep learning model to post-process reconstructed EIT thorax images to enhance its characteristics such as contrast, resolution… The model has achieved an overall accuracy of 72.43%. The resulting reconstructed images have reflected the physical changes during the breathing cycle, have deleted artifacts, and have made the shapes of the lungs and heart clearer.